Context Window Fitter

Overview

Before you call a model, we assemble your prefix, items, separator, and suffix into one canonical string and count it once with the tokenizer you choose (cl100k_base or o200k_base). We tell you whether it fits in windowTokens minus reservedOutputTokens. requiredInputTokens is the authoritative assembled count — per-item counts are informational only and must not be summed. We measure the text you send; we do not add provider chat or tool framing unless you include it in prefix, suffix, or items.

Summing per-item token counts is wrong for byte-pair encoders. We count the exact assembled string so you know before you spend on a model call.

When to use it

Scope

Non-goals

• Does not drop, reorder, or truncate items
• Does not model provider message framing
• Does not tokenize non-OpenAI encodings

Properties

Capabilities

Cost & billing

Flat rate per invocation

File handling

How we handle your data

Input schema

Output schema

API contract

curl -sS -X POST "$BASE/v1/execute" -H "Authorization: Bearer $KEY" -H 'Content-Type: application/json' -d '{"skillKey":"context.fit_window","input":{"items":[{"id":"system","text":"You are helpful."},{"id":"user","text":"Hi"}],"windowTokens":128000,"reservedOutputTokens":8000,"separator":"\n\n","tokenizer":"o200k_base"}}'

import { Squillify } from "@squillify/sdk";

const sq = new Squillify({ apiKey: process.env.SQUILLIFY_API_KEY });

const { output } = await sq.execute("context.fit_window", {"separator":"\n\n","items":[{"id":"system","text":"You are a helpful assistant."},{"id":"user","text":"Summarize the attached document."}],"reservedOutputTokens":8000,"windowTokens":128000,"tokenizer":"o200k_base"});

Errors

Other limits

Limitations

• Does not drop, reorder, or truncate items
• Does not model provider message framing
• Does not tokenize non-OpenAI encodings
• Choosing which items to keep
• Summing per-item token counts

How to call

POST /v1/execute (requires authentication)

{
  "skillKey": "context.fit_window",
  "input": {
    "separator": "\n\n",
    "items": [
      {
        "id": "system",
        "text": "You are a helpful assistant."
      },
      {
        "id": "user",
        "text": "Summarize the attached document."
      }
    ],
    "reservedOutputTokens": 8000,
    "windowTokens": 128000,
    "tokenizer": "o200k_base"
  }
}

Examples

{
  "separator": "\n\n",
  "items": [
    {
      "id": "system",
      "text": "You are a helpful assistant."
    },
    {
      "id": "user",
      "text": "Summarize the attached document."
    }
  ],
  "reservedOutputTokens": 8000,
  "windowTokens": 128000,
  "tokenizer": "o200k_base"
}

{
  "windowTokens": 128000,
  "requiredInputTokens": 13,
  "reservedOutputTokens": 8000,
  "overflowTokens": 0,
  "availableInputTokens": 120000,
  "fits": true,
  "tokenizer": "o200k_base",
  "tokenizer_version": "jtokkit-1.1.0",
  "warnings": [],
  "items": [
    {
      "id": "system",
      "independentTokenCount": 6
    },
    {
      "id": "user",
      "independentTokenCount": 7
    }
  ],
  "skill": "context.fit_window"
}

{
  "separator": "\n\n",
  "items": [
    {
      "id": "system",
      "text": "You are a helpful assistant."
    },
    {
      "id": "user",
      "text": "Summarize the attached document."
    }
  ],
  "reservedOutputTokens": 2,
  "windowTokens": 10,
  "tokenizer": "o200k_base"
}

{
  "windowTokens": 10,
  "requiredInputTokens": 13,
  "reservedOutputTokens": 2,
  "overflowTokens": 5,
  "availableInputTokens": 8,
  "fits": false,
  "tokenizer": "o200k_base",
  "tokenizer_version": "jtokkit-1.1.0",
  "warnings": [],
  "items": [
    {
      "id": "system",
      "independentTokenCount": 6
    },
    {
      "id": "user",
      "independentTokenCount": 7
    }
  ],
  "skill": "context.fit_window"
}

Recommended use cases

• Pre-flight check before an expensive model call
• Reserve output tokens and verify inputs still fit
• Measure overflow before packing or truncating

Avoid for

• Choosing which items to keep
• Summing per-item token counts

Related skills

Composable with

Safety & privacy

• Runs without calling any external service.
• Your payloads are never stored permanently.
• Telemetry captures metadata only — never your payload contents.
• We never train models on your payloads.
• No human reviews your payloads.

Tags